Project to estimate keyword score of amazon autocomplete API.
JDK - Java version 1.8.
Any Java IDE with support Maven.
Maven for build and dependecies.
###### After download the code, to build and package the application:
$ mvn clean package
###### To run the application the maven command:
$ mvn spring-boot:run
###### To get the keyword score
GET http://localhost:8080/estimate?keyword=iphone+charger
a. What assumptions did you make? I assume that if you type the first letter of a keyword and it appears in the first place in suggestion set, it means that it has the highest score possible. I also assume that if I type the entire keyword and it doesn't appear in the suggestion list it probably has the lowest score or it has not event a score at all. Additionally I also assume that for every prefix of the keyword, considering if it appears in the resulted suggestion and its position I can estimate how it influences in its score. Then I use that to implement the algorithm to estimate the keyword score.
I also assume that the keyword can have more than one word, in other words its a sentence. So "iphone x charger" is one keyword. I also assume that the order of the returned keywords from Amazon API are sorted by search volume.
b. How does your algorithm work?
Briefly it takes the input keyword and search for each prefix of this keyword in Amazon Autocomplete API. Then it processes the resulted suggestions and calculate a estimation score for the keyword.
The process to calculate the score takes the index of the first occurrence of the keyword in the resulted suggestions and send it to a score calculator. Then the score will be compute based on the parameters: the keyword length and the index of the first occurrence of the keyword. The score keyword starts with 100 and the score calculator will decrease it based on the parameters.
The calculation use the proportion of the score (100) by letter of the keyword (100 / length), since we are computing the prefixes each letter at time. Then it used the index to calculate how much of this proportion it will decrease from the score. If the index is the last in the resulted suggestion then the calculation decrease full proportion from the score. If index is the middle, 50% of the proportion, and so on.
If the index of the first occurrence of the keyword in the suggestion list, it means that following letters will not affect the result, so we finish the processing.
c. Do you think the (*hint) that we gave you earlier is correct and if so - why?
The hint is not correct. The 10 returned words are ordered by search-volume in a descendant way (the firsts elements have high search-volume than the lower elements). It doesn't make sense from amazon to send an unsorted order of keyword if the intention is to show the most searched keywords in the top of the results. Also if the order is insignificant we should see different order results for the same prefix, which is not true.
d. How precise do you think your outcome is and why?
I think the outcome is close to the Amazon score because it closely follows the real result from Amazon.com website search suggestions. This was concluded by manual tests.
If you search for "iphone charger" you got a score of 100 and it will first appear when you search for "iphone" and its prefixes, like "i", "ip", "iph", and so on. If you search for "smartphone" you will got a score of 50, which is the exact number of suggestions it is returned with this keyword (this is because the keyword has 10 letters resulting in 100 suggested results).
Those are some hypothesis that can proof that the solution is precise.
- Improve Amazon API request time
- Organize the code better using Design Patterns.